October 2026 Summaries
5 posts from QuestDB
Filter
Month:
Year:
Post Summaries
Back to Blog
QuestDB is an open-source time-series database designed for low-latency, high-ingestion workloads that can store table partitions in native format or standard Parquet while querying both seamlessly through SQL. Its example workflow uses a partitioned FX trades table, materialized OHLC views, deduplication keys, and synthetic data to show that SQL, Polars, and DuckDB produce consistent analytical results from the same Parquet-backed history. Older partitions can be converted manually or through Enterprise storage policies, reducing storage in the demonstrated case and remaining writable for late-arriving data, although such writes may require Parquet row-group rewrites and later compaction. QuestDB supports direct Parquet access by external tools, exports tables or query results to Parquet, and in Enterprise can move sealed historical partitions to object storage using Hive-style layouts readable by platforms such as Databricks, Trino, BigQuery, Snowflake, Polars, and DuckDB. The approach aims to avoid separate export pipelines and inconsistent data copies, while imposing operational constraints such as keeping QuestDB as the sole writer to managed remote files, planning cold-storage sealing beyond correction windows, and accounting for slower object-storage queries.
Oct 29, 2026
2,146 words in the original blog post.
QuestDB Enterprise 4.1, built on the QuestDB 10.1 open-source release, adds multi-tenant resource management, access auditing, identity integration, and operational monitoring for clusters shared across teams and applications. Resource groups and principal-specific memory limits can prioritize dashboards and services while constraining exploratory workloads through CPU weighting, query admission controls, queuing, and memory budgets. Audited views log each sensitive-view read with the user, outcome, timing, and resolved parameters, while object visibility limits users and client tools to tables and views for which they have permissions, and new functions expose effective and direct access grants through SQL. The release adds OIDC authentication for services, including Azure Entra managed identities and service principals, and strengthens token validation and session handling. Operational enhancements include replication, cold-storage, and certificate-health metrics in sys.metrics, longer node-level local retention for cold data, and non-blocking cold reads. Inherited QuestDB 10.1 capabilities include merge-append partitions for more efficient late-data ingestion, SUBSAMPLE for preserving chart shapes with fewer points, exact percentile functions, and faster parallel query execution. Users upgrading should review compatibility changes involving replication configuration, object visibility error responses, OIDC settings, administrative functions in views, added output columns, cold-storage requirements, and certain statements that now fail before they can suspend tables.
Oct 27, 2026
2,637 words in the original blog post.
QuestDB 10.1 is a feature-rich release for the open-source time-series database that targets lower latency, higher ingestion efficiency, and broader SQL-based analytics, while introducing more compatibility changes than a typical point release. Its opt-in merge-append partitions reduce disk rewriting for out-of-order data by appending merged partition slices and compacting them later, improving write amplification and data availability at the cost of temporary storage. New SQL capabilities include SUBSAMPLE for retaining representative real rows in charts, compression, and archives; exact discrete and continuous percentile functions with aggregate and window support; and the sys.metrics table, which stores internal metrics every second for database-native monitoring. The release also changes HORIZON JOIN behavior to preserve duplicate trade rows and correctly apply right-side predicates, accelerates GROUP BY operations over eligible joins through parallel execution, and allows list-valued DECLARE variables for filterable notebooks and dashboards. Additional improvements cover LATEST ON, indexed symbol pattern matching, time-zone sampling, large query memory use, lag and lead on symbols, WebSocket connectivity, and Web Console tools. Users are advised to review breaking changes before upgrading, particularly around HORIZON JOIN results, 32-bit INT arithmetic, interval syntax, timestamp behavior, reserved words, and platform support.
Oct 23, 2026
4,208 words in the original blog post.
QuestDB Enterprise 4.0.0 introduces a Databento connector that streams live or historical market data into QuestDB through a configuration file, replacing the Python-based ingestion approach required by the open-source edition. Using Databento’s EQUS.MINI US equities feed, the connector writes top-of-book updates and trades into automatically created, day-partitioned tables with nanosecond timestamps, order-book arrays, WAL support, and SQL-queryable schemas. The post demonstrates QuestDB SQL features including LATEST ON for current quotes, SAMPLE BY for OHLCV bars and activity analysis, and ASOF JOIN for relating trades to prevailing quotes, while Web Console notebooks provide live dashboards without requiring Grafana. Tests from a pre-market session show that retaining each update can expose spread changes, timestamp latency distributions, and millisecond-scale bursts of quote cancellations. The connector uses acknowledgement-based checkpoints, Databento intraday replay, and count-based replay deduplication to recover from restarts without gaps or accidentally discarding genuine identical records. During temporary QuestDB outages, it buffers incoming data and preserves original event timing when writes resume; replication configurations can also support failover to a promoted primary. For long-term retention, daily partitions can be transferred as Parquet files to object storage while remaining accessible through SQL, supporting both operational analysis and archival market-data workloads.
Oct 22, 2026
4,173 words in the original blog post.
QuestDB’s QWP protocol reduces time-series timestamp transfer sizes by using Gorilla encoding, a technique derived from Facebook’s Gorilla database that stores changes in timestamp intervals rather than full timestamp values. After transmitting the first two timestamps in full, QWP encodes the delta of delta between subsequent values, allowing regularly spaced timestamps to cost as little as one bit each and using variable-size bit buckets for larger timing changes. Tests described with live crypto and FX trade data show compression ranging from about 1.6x for volatile nanosecond-resolution data to roughly 57x for long, regular microsecond series, while sampled data with occasional missing intervals still achieves substantial savings. Gorilla encoding is advantageous whenever every delta-of-delta fits within a signed 32-bit range, but QWP falls back to plain 64-bit timestamps for an entire column batch if a value exceeds that limit, such as after long pauses in nanosecond data. The post also distinguishes Gorilla’s timestamp-focused implementation from Facebook’s original value compression scheme and previews QWP’s use of varints, ZigZag encoding, and bit packing for other data types.
Oct 01, 2026
2,321 words in the original blog post.